Methods

A framework for ultra-low-input spatial tissue

proteomics
Graphical abstract Authors
Anuar Makhmut, Di Qin,
Sonja Fritzsche, Jose Nimo,
Janett König, Fabian Coscia

Correspondence
fabian.coscia@mdc-berlin.de

Highlights
d Detailed protocol and guidelines for ultra-low-input spatial
tissue proteomics

d Single-cell MS-based proteomics data obtained from archival

FFPE tissue

d MS signal allows to normalize across tissue types and sample

amounts

d Cytokines and transcriptional regulators quantified in tonsil

microregions

Makhmut et al., 2023, Cell Systems 14, 1002–1014

November 15, 2023 ª 2023 Elsevier Inc.
https://doi.org/10.1016/j.cels.2023.10.003 ll
 ll

Methods
A framework for ultra-low-input
spatial tissue proteomics
Anuar Makhmut,1 Di Qin,1 Sonja Fritzsche,1 Jose Nimo,1 Janett König,1 and Fabian Coscia1,2,*
€ck-Center
1Max-Delbru for Molecular Medicine in the Helmholtz Association (MDC), Spatial Proteomics Group, Berlin, Germany
2Lead contact
*Correspondence: fabian.coscia@mdc-berlin.de
https://doi.org/10.1016/j.cels.2023.10.003

SUMMARY

Spatial proteomics combining microscopy-based cell phenotyping with ultrasensitive mass-spectrometry-

based proteomics is an emerging and powerful concept to study cell function and heterogeneity in (patho)
physiology. However, optimized workflows that preserve morphological information for phenotype discovery
and maximize proteome coverage of few or even single cells from laser microdissected tissue are currently
lacking. Here, we report a robust and scalable workflow for the proteomic analysis of ultra-low-input archival
material. Benchmarking in murine liver resulted in up to 2,000 quantified proteins from single hepatocyte con-
tours and nearly 5,000 proteins from 50-cell regions. Applied to human tonsil, we profiled 146 microregions
including T and B lymphocyte niches and quantified cell-type-specific markers, cytokines, and transcription
factors. These data also highlighted proteome dynamics within activated germinal centers, illuminating sites
undergoing B cell proliferation and somatic hypermutation. This approach has broad implications in biomed-
icine, including early disease profiling and drug target and biomarker discovery. A record of this paper’s
transparent peer review process is included in the supplemental information.

INTRODUCTION requires integrated and multimodal pipelines. We recently intro-

Cells are the functional units of organs, which fulfill essential
physiological tasks in a spatially defined manner to maintain tis-
sue integrity.1 To analyze cell dynamics in space and time,
powerful spatial genomics,2 epigenomics,3 transcriptomics,4–6
and imaging-based proteomics7,8 methods have been devel-
oped to better understand cellular and molecular drivers of
health and disease states. As proteins are the biomolecules
closest to the cellular phenotype determining cell identity and
function,9,10 spatial proteomics (SP) methods are particularly
promising for the study of human (patho)physiology. SP
methods with the single-cell resolution are dominated by tar-
geted antibody-based methods such as imaging mass cytome-
try11 (IMC) or multiplex immunofluorescence (mIF) imaging,8,12
where several dozen proteins can be analyzed at (sub)cellular
resolution. However, although such methods are well suited
for the large-scale screening of cellular phenotypes, they fall
far short of capturing the actual complexity of the cellular
proteome. It is estimated that single-cell types express more
than 10,000 unique proteins,9 which is complemented by
millions of potential proteoforms, including splice variants,
post-translational modifications (PTMs), and protein sequence
variants.10,13,14 Liquid chromatography mass spectrometry
(LC-MS)-based proteomics in contrast enables the study of pro-
teomes at an unbiased (i.e., untargeted), quantitative, and sys-
tem-wide level.9 The combination of both of these complemen-
tary proteomic approaches is therefore highly desirable but
duced deep visual proteomics (DVP),15 a new concept
combining imaging-based (fluorescence or bright field) single-
cell phenotyping with unbiased MS-based proteomics for global
proteome profiling with cell type and spatial resolution. To
realize DVP, we developed an automated laser microdissection
(LMD) workflow for the streamlined collection of nuclei, cells, or
larger regions of interest (ROIs) directly into 96- or 384-well
plates, thereby connecting whole-slide imaging and deep-
learning-based image analysis16 with ultrasensitive MS-based
proteomics.17 This allowed the profiling of as little as 100 pheno-
type-matched cells from archival tissue material while also pre-
serving detailed cell type and spatial information. Further ad-
vances in sample preparation and MS acquisition recently
pioneered the profiling of single-cell proteome heterogeneity in
cryosections of murine liver tissue,18 emphasizing the strong
spatial influence on the hepatocyte-specific proteome. Despite
these promising proof-of-concept studies, a systematic evalua-
tion and optimization of all experimental steps of IF microscopy-
guided spatial tissue proteomics is still missing. In particular, the
analysis of few or even single cells of formalin-fixed and
paraffin-embedded (FFPE) tissue collected by LMD has re-
mained elusive and relies on optimized and robust ‘‘end-to-
end’’ protocols. The successful development of such integrated
workflows could pave the way for a plethora of biomedical ap-
plications, including early disease proteome profiling studies
directly from archived patient material, where only a few cells
can be present.

1002 Cell Systems 14, 1002–1014, November 15, 2023 ª 2023 Elsevier Inc.

Methods
Here, we describe a scalable, robust, and easy-to-use proto-
col optimized for the profiling of ultra-low-input archival tissue
guided by whole-slide (IF) imaging. After benchmarking in murine
liver tissue, we applied our workflow to study the cell-type-
resolved proteome of B and T lymphocytes in different spatially
defined niches, guided by four-marker whole-slide IF imaging.
We finally provide detailed guidelines covering all aspects of
our workflow (Methods S1), from antigen retrieval (AR) and stain-
ing on LMD membrane slides, over manual or automated sample
processing, to MS data acquisition and analysis.
RESULTS
Optimizing LMD-based low-input FFPE tissue
proteomics
Light microscopy is an integral part of SP workflows that
combine LMD with MS-based proteomics19–21 (Figure 1A).
Although hematoxylin and eosin (H&E) staining protocols are
well established for tissue sections mounted on specialized
LMD membrane slides23 (Figure S1A), IF-based protocols,
which allow the sensitive detection of a higher number of
cell type and functional markers (typically 3–5 per imaging cy-
cle), require additional AR steps for epitope unmasking. The
choice of the AR method can not only influence staining and
image quality but also impacts LMD membrane integrity, tis-
sue collection efficiency, and potentially proteome coverage
of trace sample amounts. The evaluation and optimization of
AR and staining prior to MS is therefore important for ultra-
low input or even single-cell applications that strictly depend
on highly efficient tissue collection and protein extraction, as
well as near-lossless sample preparation protocols. In other
words, the most sensitive mass spectrometer can only be as
good as the upstream sample preparation workflow that de-
livers these trace peptide amounts to the instrument. We
therefore first evaluated different AR methods for their general
compatibility with whole-slide imaging on LMD slides, efficient
LMD, and ultrasensitive MS-based proteomics. We tested two
common protocols based on heat-induced epitope retrieval
(HIER) or proteolytic (i.e., pepsin-induced epitope retrieval
[PIER]) epitope retrieval and immunofluorescently stained mu-
rine liver FFPE tissue with an antibody against a ubiquitously
expressed plasma membrane marker (Na/K-ATPase). We
chose murine liver as benchmarking tissue as hepatocytes
make up 60%–80% of the liver cell mass,24 which allowed
us to obtain consistent results from serial, homogeneous tis-
sue slices, and repeated sample collections over the course
of our experiments. 5-mm-thick liver sections were mounted
on glass membrane (polyethylene naphthalate [PEN] or poly-
phenylene sulfide [PPS]) or metal frame (PPS) slides, de-paraf-
finized, and subjected to two common AR protocols (STAR
Methods) prior to antibody staining, IF microscopy, LMD,
and proteomics (Figure 1B). We included H&E stains for com-
parison, which do not rely on additional AR after de-paraffini-
zation, allowing us to directly investigate the impact of PIER
and HIER on our proteomic results. PIER was fully compatible
with both LMD slide types (glass or metal frame), facilitating
efficient LMD, but generally came at the cost of higher back-
ground staining for the tested antibody, compared with HIER
(Figure S1A). HIER occasionally resulted in membrane distor-
ll
tion of the glass-type slides toward the label end of the slide
(Figure S1B), which can impede efficient LMD collection in
this area if not prevented by the addition of glycerol.25 We
alternatively recommend tissue mounting on the side of the
slide, which is distant from the label. Frame slides on the con-
trary are more suitable for diverse biological sample types (i.e.,
tissue or cell culture) but are suboptimal for the collection of
directly connected contours, for example, for gridded sam-
pling schemes as used for the nanoPOTS approach,20 as
this can ultimately lead to loss of overall membrane integrity.
Irrespective of the choice of AR (HIER or PIER), staining tech-
nique (H&E or IF), or LMD slide type (PEN or PPS), proteomics
results from three different liver tissue amounts were highly
consistent (Figures 1C–1E). Tissue samples were processed
in 384-well low-binding plates using an MS-compatible,
organic solvent-based protocol, which included 60-min heat-
ing at 95 C for efficient formalin decrosslinking,26 sequential
Lys-C and trypsin digestion and miniaturized solid-phase
extraction. Compared with our previous DVP protocol, we
optimized our workflow for lower microliter volumes (1–2 mL)
to minimize peptide loss from surface adsorption while still be-
ing pipette-able with standard laboratory equipment. At the
same time, this also allowed the integration of robotic sample
preparation workflows (STAR Methods). In addition, we used
an optimized 15-min active nano-LC gradient (Figures S1C–
S1E) in combination with an optimal window design dia-PA-
SEF27 method on a trapped-ion mobility spectrometry (TIMS)
mass spectrometer (Bruker timsTOF SCP) for improved sensi-
tivity and sample throughput. Using DIA-NN,28 we quantified
more than 9,000 precursors and 1,700 unique proteins from
small tissue regions of approximately one to two hepatocytes
(7,500 mm3, Figures 1C, S1F, and S1G). From 50-cell sam-
ples (50,000 mm2, 5 mm thick) of H&E, HIER, and PIER-derived
samples, 4,000 proteins were consistently quantified with
excellent quantitative reproducibility (Pearson r = 0.98, Fig-
ure 1D). Furthermore, we found a 95% overlap of protein iden-
tifications across AR methods (Figure 1F), as well as nearly
identical cellular compartment proportions of the underlying
proteomes (Figure 1G), and consistent with a deep liver prote-
ome study of primary cells.22 We conclude that common FFPE
tissue preparation methods for fluorescence microscopy are
fully compatible with ultra-low input MS-based proteomics.
However, depending on the concrete application, glass mem-
brane or metal frame slides offer unique advantages and dis-
advantages, which we summarized in Figure 1H to provide
general guidelines for LMD proteomics beginners. These
data also include input from two additional LMD expert labs
and together with our detailed protocol description (Methods
S1; supplemental information), they are intended to support
the selection of the right tissue preparation and sample collec-
tion strategy for diverse SP applications.
A scalable FFPE tissue proteomics workflow allowing
single-cell analysis
Having established that common AR and staining methods are
fully compatible with LMD and ultra-low-input tissue proteomics,
we next assessed the scalability, robustness, and minimally
required sample amount of our workflow. Homogeneous areas
of murine liver tissue were collected by LMD into 384-well
Cell Systems 14, 1002–1014, November 15, 2023 1003
 ll
Methods

Figure 1. Optimizing laser microdissection-based low-input FFPE tissue proteomics

(A) Overview of the spatial tissue proteomics workflow.
(B) Tissue preparation strategy for laser microdissection-based proteomics benchmarking experiments using specialized LMD slides (metal frame or glass).
(C) Precursor and protein identifications from tissue samples processed with different staining and antigen retrieval methods. Areas of 1,562, 12,500, and
50,000 mm2 of a 5-mm thick section were laser microdissected. Averages are shown from quadruplicate measurements.
(D) Proteome correlations (Pearson’s r) of HIER, PIER, and H&E-based tissue samples.
(E) Boxplots showing coefficients of variations (CVs) of triplicate proteome measurements from the three antigen retrieval methods. Related to (D).
(F) Upset plot showing common and exclusive proteins for different antigen retrieval strategies based on 50,000 mm2 tissue samples.
(G) Protein identifications from major cellular compartments (‘‘cytosol,’’ ‘‘nucleus,’’ ‘‘plasma membrane,’’ and ‘‘extracellular region,’’ Gene Ontology Cellular
Component [GOCC]) from H&E, PIER, and HIER-treated mouse liver tissues. Percentages are the number of quantified proteins per compartment over all
quantified proteins in the corresponding sample. For comparison, a deep mouse liver proteome dataset was included22 based on non-fixed, primary cells.
(H) Summary of the laser-microdissection optimizations for low-input proteomics. Three labs assessed the applicability of glass and frame slides and rated each
category with moderate (+), good (++), and excellent (+++). The average score is shown from all three ratings. 1HIER can cause membrane distortion of glass-type
slides. 2Frame slides are more problematic for grid-based sampling schemes, the collection of many closely connected contours can lead to loss of overall
membrane integrity. 3Tested autostainers: Ventana (Roche) supported frame and glass slides, DAKO (Agilent) system supported glass slides. (A) and (B) created
with Biorender.

low-binding plates, ranging from single hepatocytes (600 mm2,
5 mm thick) to approximately 100 cells (100,000 mm2,
Figures 2A, 2B, S1F, and S1G). Proteomic results showed a
linear increase in MS2 intensity, precursor, and protein identifi-
cations from ‘‘low’’ to ‘‘high’’ tissue amounts, which, as ex-
pected, was inversely correlated with the median coefficient of
variations (CV) of protein quantifications calculated from tripli-
cate measurements of adjacent regions (Figures 2C–2E). How-
ever, even single hepatocyte contours featured low median
CVs of 20% for the 1,500–2,000 quantified proteins per contour
(Figures 2E and S2A–S2C), which we and others previously only
achieved from many thousands of cells collected from FFPE tis-
sue.30,31 It is noteworthy that CVs also included true biological
variation from known spatially defined hepatocyte heterogene-
ity,18,32 letting us conclude that reproducible single-cell prote-
ome analysis from FFPE tissue is achievable based on our

1004 Cell Systems 14, 1002–1014, November 15, 2023

 ll
Methods

Figure 2. A scalable FFPE tissue proteomics workflow allowing single-cell analysis

(A) H&E-stained mouse liver tissue section (5 mm). Collected LMD tissue samples are shown from 600 up to 100,000 mm2. Single hepatocyte contour isolation
(average size 600 mm2) was guided by the immunofluorescence signal of the plasma membrane marker Na/K-ATPase and the DAPI signal. Larger regions of
interest (ROIs) were isolated based on circular contours of pre-defined sizes ranging from 1,562 to 10,000 mm2 (2–100 cells), respectively. Hepatocyte volumes of
4,000–6,000 mm329 were used for cell count estimations of circular ROIs. Scale bars, 400 mm.
(B) Tissue inspection after collection into a 384-well low-binding plate. Scale bars, 400 mm.
(C and D) Average MS2 quantities (C), number of identified precursors (D, left), and proteins (D, right) of the collected tissue samples are shown. For all amounts,
three replicates were collected and measured. The tissue optimum is highlighted in purple and single-cell contours in orange.
(E) Boxplots showing the CVs of protein quantification across different tissue areas. CVs were calculated from triplicates of non-log-transformed data. The
boxplots define the range of the data (whiskers), 25th and 75th percentiles (box), and medians (solid line).
(F and G) Impact of different lysis buffers on proteome coverage. Tissues of 7,500, 62,500, and 250,000 mm3 of a 5-mm thick mouse liver section were profiled
using three different methods based on organic solvent only (ACN), DDM only (no ACN), or combined (DDM and ACN) protocol. For all amounts, three replicates
were collected and measured.

(legend continued on next page)

Cell Systems 14, 1002–1014, November 15, 2023 1005

 ll
workflow. We attribute the excellent quantitative reproducibility
to the combination of optimized tissue preparation, low-volume
sample processing, and highest-sensitivity MS acquisition and
analysis using an optimal window dia-PASEF scheme (STAR
Methods). Precursor and protein identifications peaked for 50–
100 cell regions (25,000–50,000 mm2), where we quantified close
to 50,000 precursors and 5,000 proteins (Table S1). Interestingly,
further increasing sample amounts resulted in lower precursor
identifications and higher median CVs, indicative of sample over-
loading (Figures 2D and 2E). Although this phenomenon could
possibly be balanced out using longer LC gradients and adjusted
trypsin amounts, thereby further improving proteome coverage,
we note that our optimized 15-min active nanoflow gradient pro-
vides an excellent compromise between single-cell sensitivity,
high proteome coverage for 50–100 cell samples, and reason-
able sample throughput of around 30–40 samples per day. Addi-
tionally, we tested the applicability of our protocol in combination
with the Bruker timsTOF Pro2, which shows a roughly 4- to 5-fold
lower total ion current (TIC) compared with the SCP instru-
ment.17 For the lowest tissue amounts measured (7,500 mm3
samples, 1–2 hepatocytes29,33), close to 1,000 proteins could
still be quantified with high quantitative reproducibility and a
somewhat similar proteome coverage was observed for the
50-cell samples (Figures S2D–S2F). We next tested if we could
further increase proteome coverage of single FFPE tissue con-
tours through the use of MS-compatible detergents. We as-
sessed the integration of n-Dodecyl-b-D-maltoside (DDM),
which was previously shown to improve proteome coverage of
low-input FFPE tissue, in particular when applied at high temper-
atures.34 Combining our acetonitrile (ACN) based protocol (10%
final concentration) with DDM (0.1% final concentration), which
likewise included 60-min controlled heating at 95 C for efficient
formalin de-crosslinking, proteome coverage of the single-cell
samples improved by 104% and 50% for precursor and protein
identifications, respectively (Figures 2F and 2G). For 62,500 (12
cells) or 250,000 mm3 (50 cells) samples, proteome coverage
was more similar between protocols. As DDM is not removed
during peptide clean-up steps, its accumulation on the analytical
column can compromise chromatographic performance over
time, which can be prevented by additional high organic solvent
washing. Our organic-solvent-based protocol performed equally
well for 50-cell contours, thus offering an excellent and cleaner
alternative to the DDM/ACN combination for ‘‘higher’’ tissue
amounts.
The unprecedented depth of our single FFPE hepatocyte con-
tours, reproducibly quantifying up to 2,000 proteins depending
on tissue thickness (Figure S2C) and with high data complete-
ness (89% complete; Figure S2G), encouraged us to further
explore these single-cell tissue proteome data. Protein levels
generally showed a high degree of concordance with higher
loading amounts (Figure S2H) and included many known hepato-
(H) Dynamic range of protein abundance for 50-cell (250,000 mm3) contours. Proteins identified in single-cell contour samples are highlighted, as well as he-
patocyte-specific markers.22 A minimum of two quantified values per quadruplicate measurement was required for the single-contour samples.
(I) Histogram of log10 protein intensities obtained from 600 and 50,000 mm2 samples. Proteins identified in single cells cover 2–3 orders of magnitude from the top
abundant fraction of the liver proteome.
(J) Reactome and KEGG pathway coverage for single-cell and 50-cell samples. Values show the number of proteins quantified per pathway. For comparison, a
deep (>10,000 proteins) mouse liver dataset22 was included. A minimum of two quantified values from quadruplicate measurements was required for single cells
and three values from triplicates for the 50-cell samples.
1006 Cell Systems 14, 1002–1014, November 15, 2023
Methods
cyte-specific markers distributed over a dynamic range of
approximately three orders of magnitude (Figures 2H and 2I).
Proteins involved in ‘‘housekeeping’’ cell functions were quanti-
fied at a similar depth of analysis as the 50-cell samples, despite
the approximately 33-fold lower total MS2 signal, and, remark-
ably, compared with the deep mouse liver study covering more
than 10,000 proteins22 (Figure 2J). For example, we quantified
70 of 81 ribosomal proteins, 29 of 44 involved in fatty acid meta-
bolism and 22 of 30 tricarboxylic acid (TCA) cycle-related pro-
teins (Table S2). Interestingly, although housekeeping functions
such as ribosomal proteins were more stably expressed, cell-
metabolism-related proteins showed higher variation in the sin-
gle-cell contours compared with the 50-cell contours, likely due
to proteome averaging (Figure S2I). This is in line with a recent
liver study revealing strong metabolic differences in single hepa-
tocytes along the liver zonation axis.18 For lower abundant path-
ways, for example, insulin signaling or major histocompatibility
complex (MHC)-2 antigen presentation (12 vs. 107 and 25 vs.
92 detected proteins, respectively) (Figure 2J), a higher discor-
dance in detected proteins was observed between the low and
high tissue amounts, making the comprehensive single-cell anal-
ysis of these pathways only achievable with further improved
workflow sensitivity, or alternatively, by pooling phenotype-
matched cells, as we conceptualized recently.15 We conclude
that reproducible single-cell-based FFPE tissue profiling is
achievable using optimized sample processing and ultrasensi-
tive LC-MS workflows revealing important insights into cell
identity and function. The scalability of our workflow also enabled
the spatially resolved quantification of 5,000 proteins from
50-cell regions, capturing a substantial fraction of the cell-type-
specific proteome, thereby complementing single-cell-based
analyses.
Optimizing sample input across tissue and cell types
Based on our tissue dilution experiment in the murine liver, we
empirically determined the optimal tissue amount for the highest
proteome coverage of small tissue areas using a 15-min active
nanoflow gradient combined with dia-PASEF on the Bruker tim-
sTOF SCP. We next addressed how this liver optimum translated
to other tissue and cell types and how the recorded MS readout
could be exploited to normalize sample loading from tissue to
tissue or cell type to cell type. Such adjustments are of particular
importance for ultra-low sample amounts, which are not
amenable to peptide concentration measurements routinely
used prior to MS bulk analysis. In addition, various factors can
affect the TIC derived from different tissue specimens, including
sample-related sources of variability such as tissue archival
time, which can affect the retrieval of lower abundant proteins,35
or biologically due to protein abundance differences across tis-
sue and cell types.36 In any case, normalizing and adjusting sam-
ple amounts is hence particularly important for low input tissue
 ll
Methods

Figure 3. Optimizing sample input across tissue and cell types

(A) Immunofluorescence whole-slide image of a 5-mm-thick tonsil tissue section stained for CD3 (T cells), CD19 (B cells), and DNA (DAPI). Scale bars, 500 mm.
(B) Magnifications of exemplary B and T cell enriched regions used for laser microdissection and proteomics profiling. Scale bars, 25 mm.

(legend continued on next page)

Cell Systems 14, 1002–1014, November 15, 2023 1007

 ll
Methods

proteomics and should be carefully assessed. We hypothesized
that a simple normalization strategy using the total excised tis-
sue volume (microdissected area 3 section thickness) across
specimens should be a poor estimator of the optimal sample
amount. To test this, we analyzed a second tissue type, the hu-
man FFPE tonsil, which as a secondary lymphoid organ is impor-
tant for the development of immune tolerance and adaptive im-
mune functions and comprises B and T cell subsets, as well as
other immune and non-immune-related cell types.37 In murine
liver, 50–100 cell contours (125,000–250,000 mm3) yielded the
highest proteome depth and lowest median protein CVs
(Figures 2C–2E and S3A–S3C). Analyzing the same tissue
amounts obtained from tonsil, focusing on T and B cell enriched
microregions after IF imaging (Figures 3A and 3B), revealed
clearly different sampling optima (Figure 3C), despite a similar to-
tal number of quantified proteins per experiment (5,000–5,500
proteins; Table S3). Based on the MS2 signal (sum of MS2 quan-
tities of all peaks) and total quantity (sum of MS2 quantities of
identified precursors) retrieved from the DIA-NN output, we esti-
mated that murine liver on average resulted in roughly 3-fold
(2.83) higher MS2 signal compared with tonsil (Figure 3D), likely
due to different protein abundances in these two organs (Fig-
ure 3E). Notably, the MS2 signal derived from the liver reference
sample was consistent over the tested time frame of 6 months
and independent of section thickness (Figure S3D). In other
words, a 5-mm-thick contour of 25,000 mm2 resulted in an almost
identical MS2 signal as a 10-mm-thick contour of 12,500 mm2.
The optimal tissue amount for tonsil was 350,000–700,000 mm3
instead, allowing us to quantify 35,000 precursors and
5,000 proteins in the 15-min dia-PASEF measurement (Fig-
ure 3C), and beyond which no further increase in proteome depth
was apparent. In fact, we observed lower identification rates
beyond this saturation point, similar to our findings in liver
(Figures 2C–2E), which we mostly attribute to increased tryptic
miscleavage (Figure S3E), as there was no sign of TIMS satura-
tion for the tested tissue amounts. However, further increased
peptide amounts could result in precursor-intensity-dependent
TIMS cell fragmentation, emphasizing the need to carefully
assess and normalize sample loading. This prompted us to
test whether these tissue saturation points obtained from the
liver and tonsil data could be exploited to predict the best tissue
sampling amount from ‘‘single-shot’’ measurements alone. This
simulates a scenario at the beginning of a spatial tissue prote-
omics study, where a priori proteomics information for a new tis-
sue type is lacking. The accurate prediction of the optimal tissue
amount could hence save time and resources and maximize
data output. To this end, we laser microdissected replicates of
small 60,000 mm3 samples (6,000 mm2 of a 10 mm slice) of a
consecutive tonsil tissue section, such that the obtained inten-
sity was in the linear range of the mass spectrometric signal
(>50,000 mm3 tissue; Figure 3D) and compared the recorded
MS2 precursor quantity with our pre-determined optimum (MS
signal = 6–8 3 1011). To minimize variability from different extra-
cellular matrix compositions, we focused on homogeneous tis-
sue regions of high cellularity (Figure 3B), similar to the liver titra-
tion experiment (Figure S1G). This way, we extrapolated that
450,000–753,000 mm3 would be the optimal sampling amount
for this tonsil tissue, in excellent agreement with our measure-
ments (Figure 3C). We note that this simple normalization strat-
egy is strictly dependent on the employed LC-MS setup and
therefore requires empirical data to begin with but generally
recommend to perform such ‘‘survey’’ experiments to pinpoint
the optimal tissue amount needed for a specific SP research
question. In our example, murine liver served as highly consis-
tent reference tissue to predict the ideal tonsil amount; however,
other tissue types should be equally suited for such normaliza-
tion. Moreover, this strategy can prevent TIMS overloading,
which can negatively affect proteome coverage and quantifica-
tion. As a positive byproduct, the acquired raw files of the titra-
tion data can also be used to create project-specific refined
spectral libraries (STAR Methods), thereby drastically speeding
up subsequent DIA-NN searches.
As the size of the total dissected tissue area (or the number of
collected single-cell contours per sample), which determines the
obtained spatial resolution, is inversely correlated with proteome
coverage (Figures 2C and 2D), such survey measurements can
also help to find a good balance between these two parameters
in the context of the specific research question. To demonstrate
this exemplarily, we further analyzed our tonsil proteome data,
which in total covered more than 5,000 proteins distributed
over four orders of magnitude (Figure 3F). Projecting the different
tissue dilution measurements onto the measured dynamic range
of protein abundance showed that B or T cell-specific regions of
only 8,750 mm2 (5-mm-thick section) were already sufficient to
quantify many key players of immune cell signaling, cell-type-
specific markers, cytokines, and even transcription factors
(e.g., STAT1, IRF3, IFNGR1, CD8A, CD19, Ki-67, LAG3, inter-
leukin [IL]-16, and IL-18; Figures 3B, 3G, 3H, S3F, and S3G) at
a spatial resolution of 70–100 mm (center-to-center; Figure 3A).
Consequently, our pipeline should allow the analysis of spatially
resolved proteomes of various B and T cell niches from regions
of as little as 4,000 mm2 (10-mm-thick section).

(C) Bar plots of MS2 intensities (left), precursors (middle) and proteins quantified (right) obtained from increasing amounts of tonsil tissue. For all amounts, six
replicates were collected and measured. Note, the tissue-specific sampling optimum was reached at 350,000–700,000 mm3, beyond which identifications
dropped again.
(D) Comparison of liver and tonsil tissue dilution data. MS2 quantities are plotted against increasing (log2 transformed) tissue amounts (volume in mm3). Based on
the empirically determined MS2 intensity optimum (MS2 quantity = 6–8E11 for liver and tonsil tissues), 2.8-fold more tonsil tissue was sampled to reach the same
MS2 quantity. For tonsil and liver tissue samples, data show averages from minimum six and five replicates per group, respectively.
(E) Cumulative protein intensities of liver and tonsil tissue proteomes ranked from the highest to the lowest abundant protein. Density plot shows protein intensity
distributions for both tissues. Top 20 proteins of each tissue are shown on the right.
(F) Dynamic range of protein abundance from different amounts of tonsil tissue.
(G) Histogram of protein intensities of the of 8,750 mm2 tonsil tissue sample (43,750 mm3 in volume). Proteins highlighted in pink belong to an RNA-seq-based
signature of tonsil germinal centers (GSE12845), including known immune and cell-type-specific markers.
(H) Unsupervised hierarchical clustering of small (6,000 mm2 3 10 mm) B cell, T cell, and mixed B/T cell regions, related to (B). Z scored protein levels indicate
upregulated (red) or downregulated (blue) proteins.

1008 Cell Systems 14, 1002–1014, November 15, 2023

 ll
Methods

(legend on next page)

Cell Systems 14, 1002–1014, November 15, 2023 1009

 ll
Methods

Spatially and cell-type-resolved proteomics of human
tonsil tissue
Encouraged by our tonsil titration data, which revealed that known
immune cell regulators, cytokines, and transcription factors were
quantified from tissue regions of as little as 8,750 mm2
(43,750 mm3 in volume), we systematically investigated the impact
of the spatial location on the B and T cell specific proteomes. Hu-
man tonsil represents a prime example of tissues that are orga-
nized into distinct microanatomical compartments to fulfill diverse
biological functions critical for adaptive immunity. Following anti-
gen encounter of naive B cells within the follicle, secondary (acti-
vated) follicles are rapidly formed as production centers of anti-
gen-specific B cells. Following activation, B cells undergo cycles
of maturation and selection within newly formed germinal centers
(GCs), ultimately giving rise to highly antigen-specific antibody-
secreting plasma cells and memory B cells, the key players of hu-
moral immune response.38
We first mounted a 10-mm-thick tonsil section obtained from a
patient who underwent bilateral tonsillectomy on a metal frame
LMD slide and performed four-color IF whole-slide imaging to
detect B cells (CD19), T cells (CD3), epithelium (EP) (pan-CK),
and DNA (DAPI) (Figures 4A and 4B). Using the open-source image
analysis software QuPATH,39 we selected a total of 146 microre-
gions for automated LMD (STAR Methods) and quantitative prote-
omics (Table S4). Based on the well-defined tissue architecture of
the human tonsil (Figure 4A), we included small circular regions of
4,000 mm2 isolated from primary and secondary B cell follicles,
including subregions of dark, light, and gray GC B cell niches, naive
mantel-zone-derived B cells, various interfollicular T cell zones,
and squamous cell EP (Figures 4A–4C). On average, we quantified
1,952 proteins per sample (Figure S4C) and 3,334 proteins in total.
After data filtering and missing value imputation (STAR Methods),
proteomes clearly separated by microanatomical regions domi-
nated by the distinct cell types (Figure 4D). Known cell type
markers such as CD3D (T cell marker), CD19 (B cell marker), and
CDH1 (E-Cadherin, epithelial marker) were highest in the expected
sample groups (Figure 4E), confirming the high specificity of our
proteome data. GCs showed clear signs of increased proliferation
(e.g., Ki-67, PCNA) and DNA repair (e.g., TOP2A), indicative of
active sites of B cell expansion and somatic hypermutation.40
T cell zones instead featured high levels of the STAT1 transcription
factor as well as T cell modulating cytokines such as IL-16 (Fig-
ure 4F). Moreover, the global comparison of all T cells (n = 34) vs.
B cell-specific proteomes (mantle zone [MZ], n = 36) revealed
many bona fide B and T cell markers among the top regulated pro-
teins (e.g., CD22, CD3E, CD72, and CD5) and potentially many
other less-characterized ones (Figure 4G). We next focused on
different B cell niches to assess if our spatially resolved data
captured known functional differences of spatially defined B cell
zones. Naive (MZ) and activated (GC) B cells showed strong prote-
ome differences (Figure 4H), indicative of their unique biological
functions. Although MZ B cells were characterized by higher
metabolic activity (e.g., TCA cycle and fructose and mannose
metabolism), senescence signatures, and phosphatidylinositol
signaling, activated GC B cells showed strong replication and
DNA repair signatures (Figures 4H and 4I; Table S4), in line with
their known biological function. Encouraged by this, we next as-
sessed whether our quantitative data even separated spatially
defined GC sub-compartments of dark (sites undergoing active
B cell proliferation and somatic hypermutation) vs. light (site of B
cell selection) zones (Figures 4J and 4K). Clearly, dark-zone-
derived B cells featured higher replication and DNA damage
response profiles compared with the light zone (Figures 4L and
4M). The higher expression of the T cell receptor subunit CD3D
in light zones on the contrary illuminated the presence of T helper
cells, important for B cell selection.40 We confirmed this finding by
assessing our imaging data, which indeed revealed significantly
higher CD3 signals in the light GC regions, which was not the
case for the B cell marker CD19 (Figure S4D).
In summary, our data delineated how the robust microscopy-
guided ultra-low-input tissue proteomics workflow introduced
here can be applied to study cell type and spatially resolved pro-
teomes in health and disease based on readily accessible
archival FFPE specimens.

Figure 4. Spatially and cell-type-resolved proteomics of human tonsil tissue

(A) Immunofluorescence whole-slide image of a 10-mm thick tonsil tissue section stained for CD3 (T cells), CD19 (B cells), pan-CK (epithelium), and DNA (DAPI).
Scale bars, 250 mm.
(B) Magnifications of the exemplary epithelium (EP), germinal center (GC), mantle zone (MZ), and T cell enriched regions used for laser microdissection and
proteomic profiling. Scale bars, 20 mm.
(C) Sample collection strategy and the total number of samples for each tissue region.
(D) 3D principal-component analysis (PCA) of 146 samples based on 2,235 protein groups after data filtering and imputation.
(E and F) Log2-protein levels of cell-type-specific and functional markers quantified in different tissue regions. Black dots indicate average values for each group.
Asterisks indicate ANOVA p values of <0.001 after data imputation.
(G) Volcano plot of the pairwise proteomic comparison between the B cell zone (mantle zone) and T cell zone. Cell type-specific markers are highlighted in green
and turquoise (two-sided t test, false discovery rate [FDR] < 0.05).
(H) Volcano plot of the pairwise proteomic comparison between mantle zone and germinal center samples. Cell type-specific markers are highlighted in green and
blue (two-sided t test, FDR < 0.05).
(I) Pathway enrichment analysis (Reactome and KEGG) based on t test difference between mantel zone and germinal center samples. Selected pathways with a
Benjamin-Hochberg FDR < 0.05 are shown.
(J) ROIs used for proteomic profiling of a secondary follicle region. Dark (red), gray (orange), and light (yellow) germinal center zones were selected for proteomic
profiling. Mantel zone regions (light gray) are shown additionally. Scale bars, 50 mm.
(K) PCA of dark, gray, and light zone proteomes. Point concentration ellipses are shown for each group with a 95% confidence.
(L) Boxplots of relative protein levels (Z score) for selected markers. The boxplots define the range of the data (whiskers), 25th and 75th percentiles (box), and
medians (solid line). Asterisks indicate two-sided t test p values (dark vs. light zone) of p < 0.001.
(M) Unsupervised hierarchical clustering of ANOVA significant proteins (permutation-based FDR < 0.05) from dark, gray, and light zone samples. Related to
(J) and (K). Heatmap shows relative protein levels (Z score) of upregulated (yellow) and downregulated proteins (black). *KRT78 is marked as potential
contaminant.

1010 Cell Systems 14, 1002–1014, November 15, 2023


Methods
DISCUSSION
Spatial tissue proteomics connecting microscopy-based cell
phenotyping with LMD guided MS-based proteomics is an
emerging discovery concept for the study of cell function and
heterogeneity in health and disease. Our group recently co-
developed deep visual proteomics, an approach that combines
high-parametric imaging and machine-learning-based single-
cell phenotyping to guide precise tissue sampling for ultrasensi-
tive LC-MS analysis. This enabled the profiling of as little as 100
tissue cells per sample to a depth of 3,000–5,000 proteins,
dependent on the tissue and cell type of interest. However, the
flexible and highly modular design of the DVP pipeline, enabling
the profiling of single or few cells on one hand, or hundreds of
phenotype-matched cells for deeper proteome interrogation
(i.e., 5,000 proteins or more) on the other hand, also necessitates
carefully designed tissue benchmarking experiments and
detailed guidelines to extract most information for diverse
biomedical applications. With this in mind, we here introduce,
benchmark, and apply an optimized end-to-end workflow, start-
ing from AR comparisons for immunostaining and microscopy,
over guidelines for best-practice tissue collection by LMD, to-
ward an optimal nanoflow dia-PASEF LC-MS scheme for high-
est-sensitivity MS-based proteomics. For the first time, we pro-
vide evidence that robust ultra-low-input proteomics (few- or
even single-excised cells) from FFPE tissue slices is achievable
and fully compatible with conventional AR and four-marker IF im-
aging protocols, paving the way for higher-plex IF combinations
in the near future. At the same time, our flexible 384-well design
makes the protocol easily adaptable to robotic workflows for
further protocol automation. We find that low microliter volumes
(2 mL) for sample preparation are sufficient to achieve true sin-
gle-cell sensitivity, which makes this workflow principally acces-
sible to any laboratory. On the Bruker timsTOF SCP, this allowed
us to quantify up to 2,000 proteins from single hepatocyte con-
tours and around 5,000 proteins from 50-cell regions, demon-
strating excellent workflow scalability. On the timsTOF Pro2 in-
strument, this translated to nearly 1,000 high-confidence
proteins from 1 to 2 hepatocytes (1,500 mm2 regions) and a
somewhat similar proteome depth for the 50-cell samples,
further emphasizing the broad applicability of our protocol. The
data from single-excised hepatocytes also revealed that a
25-mm spatial resolution is principally achievable for tissue
types such as liver, on par with the spatial resolution of state-
of-the-art spatial transcriptomics.41,42 These data also show
that highly abundant housekeeping proteins, such as ribosomal
proteins, are quantifiable at a depth and precision similar to
much higher sampling amounts, for example, from large-scale
bulk measurements. Our data also show that many metabolic
pathways (e.g., TCA cycle or fatty acid metabolism) are likewise
amenable to single-cell-based proteomic analysis, as also
shown recently.18 Notably, housekeeping related proteins are
often poorly correlated to mRNA abundances,10,43 making these
pathways particularly attractive for early applications of single-
cell-based tissue proteomics.
We also provide guidance on how to empirically determine the
optimal tissue amount to achieve the highest proteome coverage
while avoiding instrument overloading, which is particularly perti-
nent for single-cell sensitivity MS setups. We introduce a simple
ll
normalization strategy using the extracted MS2 signal directly
from the search results to find the optimal tissue amount for liter-
ally any low-input spatial tissue proteomics experiment. This
strategy can also be applied to normalize qualitative sample dif-
ferences, which are often observed for archival specimens due
to, for example, varying archival times.31 The importance of
this can be illustrated through the eyes of a pathologist. For
the histomolecular analysis of cancer progression states, sam-
ples are typically distributed over several FFPE tissue blocks
(for example pre-cancer, primary tumor, and metastasis), poten-
tially collected over many years. Survey experiments, as outlined
here using the example of human tonsil tissue, could guide the
best sampling strategy for deep and reproducible LMD-assisted
proteomics.
However, although the latest generation MS instruments
feature excellent sensitivity when combined with optimized sam-
ple preparation workflows, such as the one introduced here, MS
throughput is still a major bottleneck for tissue sections that are
often larger than 1 cm2. Isobaric or non-isobaric multiplexing
strategies generally offer good alternatives to label-free based
methods further increasing sample throughput,44–46 but they
are still not sufficient to deal with this tremendous analytical
bottleneck. One can estimate that for the profiling of one 1 3
1 cm tissue section, sampled by non-overlapping 50 3 50 mm
squares, amounts, which we here show in liver and tonsil tissues
to be amenable to the reproducible quantification of 2,000–3,000
proteins, 40,000 measurements would be required for gridded
whole-slide sampling schemes. This is far beyond the reach of
current LC-MS setups, which typically analyze 20–50 tissue pro-
teomes per day. Instead, the integration of whole-slide IF imag-
ing for detailed cell and cellular neighborhood phenotyping al-
lows us to prioritize cells and ROIs subjected to global
proteome analysis, thereby offering a powerful, cost-effective,
and accessible spatial profiling strategy. We illustrate this in
tonsil tissue, where we use four-marker whole-slide IF imaging
to guide the sampling of over 140 microregions per single batch,
covering naive and activated B cell niches, interfollicular T cell
zones and squamous cell EP. From only 63 3 63 mm regions
(4,000 mm2 regions), we quantified spatially resolved proteomes
of activated GC niches, illuminating sites of clonal B cell expan-
sion and somatic hypermutation. Intriguingly, based on these
1,500–3,000 protein measurements per microregion, we quanti-
fied key players of immune cell function including cytokines and
transcriptional regulators, emphasizing the power of our ‘‘biolog-
ical fractionation’’ strategy to dig deep into the cell type and
spatially resolved tissue proteome.
In conclusion, we here provide a scalable and optimized
framework for MS-based spatial tissue proteomics of ultra-
low-input archival specimens combining high-content imaging,
LMD, and ultrasensitive MS.
STAR+METHODS
Detailed methods are provided in the online version of this paper
and include the following:
d KEY RESOURCES TABLE
d RESOURCE AVAILABILITY
B Lead contact
Cell Systems 14, 1002–1014, November 15, 2023 1011
 ll
Methods
B Materials availability sequencing. Cell 172, 205–217.e12. https://doi.org/10.1016/j.cell.2017.
B Data and code availability 12.007.
d EXPERIMENTAL MODEL AND SUBJECT DETAILS 3. Deng, Y., Bartosovic, M., Kukanja, P., Zhang, D., Liu, Y., Su, G., Enninful,
B Mouse experiments and organ harvesting A., Bai, Z., Castelo-Branco, G., and Fan, R. (2022). Spatial-CUT&Tag:
B Human tissue samples spatially resolved chromatin modification profiling at the cellular level.
Science 375, 681–686. https://doi.org/10.1126/science.abg7216.
d METHOD DETAILS
4. Marx, V. (2021). Method of the Year: spatially resolved transcriptomics.
B Hematoxylin-eosin staining
Nat. Methods 18, 9–14. https://doi.org/10.1038/s41592-020-01033-y.
B Immunofluorescence staining
5. Longo, S.K., Guo, M.G., Ji, A.L., and Khavari, P.A. (2021). Integrating single-
B High-resolution microscopy
cell and spatial transcriptomics to elucidate intercellular tissue dynamics. Nat.
B Image analysis and contour export for laser microdis-
Rev. Genet. 22, 627–644. https://doi.org/10.1038/s41576-021-00370-8.
section
6. He, S., Bhatt, R., Brown, C., Brown, E.A., Buhr, D.L., Chantranuvatana, K.,
B Laser microdissection Danaher, P., Dunaway, D., Garrison, R.G., Geiss, G., et al. (2022). High-
B Sample preparation for LC-MS analysis plex multiomic analysis in FFPE at subcellular level by spatial. Mol.
B Peptide clean-up with C-18 tips Imaging 40, 1794–1806. https://doi.org/10.1101/2021.11.03.467020.
B Liquid chromatography–mass spectrometry (LC – MS) 7. Baharlou, H., Canete, N.P., Cunningham, A.L., Harman, A.N., and Patrick,
analysis E. (2019). Mass cytometry imaging for the study of human diseases—ap-
B Proteomics raw data analysis plications and data analysis strategies. Front. Immunol. 10, 2657. https://
d QUANTIFICATION AND STATISTICAL ANALYSIS doi.org/10.3389/fimmu.2019.02657.
8. Schu €rch, C.M., Bhate, S.S., Barlow, G.L., Phillips, D.J., Noti, L., Zlobec, I.,
Chu, P., Black, S., Demeter, J., McIlwain, D.R., et al. (2020). Coordinated
SUPPLEMENTAL INFORMATION
cellular neighborhoods orchestrate antitumoral immunity at the colorectal
cancer invasive front. Cell 182, 1341–1359.e19. https://doi.org/10.1016/j.
Supplemental information can be found online at https://doi.org/10.1016/j.
cell.2020.07.005.
cels.2023.10.003.
9. Aebersold, R., and Mann, M. (2016). Mass-spectrometric exploration of
ACKNOWLEDGMENTS proteome structure and function. Nature 537, 347–355. https://doi.org/
10.1038/nature19949.
We would like to thank our colleagues at the Max Delbru €ck Center (MDC) for 10. Buccitelli, C., and Selbach, M. (2020). mRNAs, proteins and the emerging
their support and fruitful discussions. In particular, we thank Ulrike Stein for principles of gene expression control. Nat. Rev. Genet. 21, 630–644.
her support to perform the mouse liver experiments. We thank Philipp Mertins https://doi.org/10.1038/s41576-020-0258-4.
and Simon Haas for their critical feedback on the manuscript and Christian 11. Jackson, H.W., Fischer, J.R., Zanotelli, V.R.T., Ali, H.R., Mechera, R.,
Sommer for MS support. We also thank Andreas Mund (Center for Protein Soysal, S.D., Moch, H., Muenst, S., Varga, Z., Weber, W.P., et al. (2020).
Research, University of Copenhagen) and Lisa Schweizer (MPI of Biochem- The single-cell pathology landscape of breast cancer. Nature 578,
istry, Munich) for their input and feedback on the LMD membrane compari- 615–620. https://doi.org/10.1038/s41586-019-1876-x.
sons. We thank Simon Schallenberg (Charité Pathology, Berlin) for his help
12. Lin, J.R., Izar, B., Wang, S., Yapp, C., Mei, S., Shah, P.M., Santagata, S.,
with the tonsil tissue experiments. Furthermore, we acknowledge the MDC
and Sorger, P.K. (2018). Highly multiplexed immunofluorescence imaging
technology platforms ‘‘Proteomics’’ and ‘‘Advanced Light Microscopy’’ for
of human tissues and tumors using t-CyCIF and conventional optical mi-
their great support. All authors acknowledge support by the Federal Ministry
croscopes. eLife 7, 1–46. https://doi.org/10.7554/eLife.31657.
of Education and Research (BMBF), as part of the National Research Initiatives
for Mass Spectrometry in Systems Medicine, under grant agreement No. 13. Sharma, K., D’Souza, R.C.J., Tyanova, S., Schaab, C., Wisniewski, J.R.,
161L0222. Cox, J., and Mann, M. (2014). Ultradeep human phosphoproteome reveals
a distinct regulatory nature of Tyr and ser/Thr-based signaling. Cell Rep. 8,
1583–1594. https://doi.org/10.1016/j.celrep.2014.07.036.
AUTHOR CONTRIBUTIONS
14. Sinitcyn, P., Richards, A.L., Weatheritt, R.J., Brademan, D.R., Marx, H.,
Conceptualization, A.M. and F.C.; methodology, A.M., D.Q., J.K., J.N., and Shishkova, E., Meyer, J.G., Hebert, A.S., Westphall, M.S., Blencowe,
F.C.; experiments, A.M., D.Q., S.F., J.K., and F.C.; data analysis, A.M. and B.J., et al. (2023). Global detection of human variants and isoforms by
F.C.; figures, A.M. and F.C.; supervision, F.C.; funding acquisition, F.C.; writing deep proteome sequencing. Nat. Biotechnol. https://doi.org/10.1038/
the original draft, F.C. All the authors reviewed and edited the manuscript. s41587-023-01714-x.
15. Mund, A., Coscia, F., Kriston, A., Hollandi, R., Kovács, F., Brunner, A.-D.,
DECLARATION OF INTERESTS Migh, E., Schweizer, L., Santos, A., Bzorek, M., et al. (2022). Deep Visual
proteomics defines single-cell identity and heterogeneity. Nat. Biotechnol.
The authors declare no competing interests.
40, 1231–1240. https://doi.org/10.1038/s41587-022-01302-5.

Received: June 1, 2023 16. Hollandi, R., Szkalisity, A., Toth, T., Tasnadi, E., Molnar, C., Mathe, B.,
Revised: August 3, 2023 Grexa, I., Molnar, J., Balind, A., Gorbe, M., et al. (2020). nucleAIzer: A
Accepted: October 6, 2023 parameter-free deep learning framework for nucleus segmentation using
Published: October 30, 2023 image style transfer. Cell Syst. 10, 453–458.e6. https://doi.org/10.1016/j.
cels.2020.04.003.
REFERENCES 17. Brunner, A.D., Thielert, M., Vasilopoulou, C., Ammar, C., Coscia, F., Mund,
A., Hoerning, O.B., Bache, N., Apalategui, A., Lubeck, M., et al. (2022).
1. Palla, G., Fischer, D.S., Regev, A., and Theis, F.J. (2022). Spatial compo- Ultra-high sensitivity mass spectrometry quantifies single-cell proteome
nents of molecular tissue biology. Nat. Biotechnol. 40, 308–318. https:// changes upon perturbation. Mol. Syst. Biol. 18, e10798. https://doi.org/
doi.org/10.1038/s41587-021-01182-1. 10.15252/msb.202110798.
2. Casasent, A.K., Schalck, A., Gao, R., Sei, E., Long, A., Pangburn, W., 18. Rosenberger, F.A., Thielert, M., Strauss, M.T., Ammar, C., Ma €dler, C.,
Casasent, T., Meric-Bernstam, F., Edgerton, M.E., and Navin, N.E. (2018). Schweizer, L., Metousis, A., Skowronek, P., Wahle, M., Semenova, A.,
Multiclonal invasion in breast tumors identified by topographic single cell et al. (2023). Spatial single-cell mass spectrometry defines zonation of the

1012 Cell Systems 14, 1002–1014, November 15, 2023


Methods
hepatocyte proteome author list. Nat Methods 20, 1530–1536. https://doi.
org/10.1038/s41592-023-02007.
19. Wisniewski, J.R., Ostasiewicz, P., and Mann, M. (2011). High recovery
FASP applied to the proteomic analysis of microdissected formalin fixed
paraffin embedded cancer tissues retrieves known colon cancer markers.
J. Proteome Res. 10, 3040–3049. https://doi.org/10.1021/pr200019m.
20. Piehowski, P.D., Zhu, Y., Bramer, L.M., Stratton, K.G., Zhao, R., Orton,
D.J., Moore, R.J., Yuan, J., Mitchell, H.D., Gao, Y., et al. (2020).
Automated mass spectrometry imaging of over 2000 proteins from tissue
sections at 100-mm spatial resolution. Nat. Commun. 11, 8. https://doi.org/
10.1038/s41467-019-13858-z.
21. Eckert, M.A., Coscia, F., Chryplewicz, A., Chang, J.W., Hernandez, K.M.,
, I., et al. (2019).
Pan, S., Tienda, S.M., Nahotko, D.A., Li, G., Blazenovic
Proteomics reveals NNMT as a master metabolic regulator of cancer-
associated fibroblasts. Nature 569, 723–728. https://doi.org/10.1038/
s41586-019-1173-8.
22. Azimifar, S.B., Nagaraj, N., Cox, J., and Mann, M. (2014). Cell-type-
resolved quantitative proteomics of murine liver. Cell Metab. 20, 1076–
1087. https://doi.org/10.1016/j.cmet.2014.11.002.
23. Emmert-Buck, M.R., Bonner, R.F., Smith, P.D., Chuaqui, R.F., Zhuang, Z.,
Goldstein, S.R., Weiss, R.A., and Liotta, L.A. (1996). Laser capture micro-
dissection. Science 274, 998–1001. https://doi.org/10.1126/science.274.
5289.998.
24. Ben-Moshe, S., and Itzkovitz, S. (2019). Spatial heterogeneity in the
mammalian liver. Nat. Rev. Gastroenterol. Hepatol. 16, 395–410. https://
doi.org/10.1038/s41575-019-0134-x.
25. Nordmann, T.M., Schweizer, L., Metousis, A., and Thielert, M. (2023).
Rahbek-gjerdrum, L.M., Stadler, P, and Bzorek, M. A standardized and
reproducible workflow for membrane glass slides in routine histology
and spatial proteomics. https://doi.org/10.1101/2023.02.20.529255.
26. Kawashima, Y., Kodera, Y., Singh, A., Matsumoto, M., and Matsumoto, H.
(2014). Efficient extraction of proteins from formalin-fixed paraffin-
embedded tissues requires higher concentration of tris ( hydroxymethyl )
aminomethane. Clin. Proteomics 11, 4. https://doi.org/10.1186/1559-
0275-11-4.
27. Meier, F., Brunner, A.D., Frank, M., Ha, A., Bludau, I., Voytik, E., Kaspar-
Schoenefeld, S., Lubeck, M., Raether, O., Bache, N., et al. (2020).
diaPASEF: parallel accumulation–serial fragmentation combined with
data-independent acquisition. Nat. Methods 17, 1229–1236. https://doi.
org/10.1038/s41592-020-00998-0.
28. Demichev, V., Messner, C.B., Vernardis, S.I., Lilley, K.S., and Ralser, M.
(2020). DIA-NN: neural networks and interference correction enable
deep proteome coverage in high throughput. Nat. Methods 17, 41–44.
https://doi.org/10.1038/s41592-019-0638-x.
29. Hammad, S., Hoehme, S., Friebel, A., Von Recklinghausen, I., Othman, A.,
Begher-Tibbe, B., Reif, R., Godoy, P., Johann, T., Vartak, A., et al. (2014).
Protocols for staining of bile canalicular and sinusoidal networks of human,
mouse and pig livers, three-dimensional reconstruction and quantification of
tissue microarchitecture by image processing and analysis. Arch. Toxicol.
88, 1161–1183. https://doi.org/10.1007/S00204-014-1243-5/FIGURES/9.
30. Guo, T., Kouvonen, P., Koh, C.C., Gillet, L.C., Wolski, W.E., Röst, H.L.,
Rosenberger, G., Collins, B.C., Blum, L.C., Gillessen, S., et al. (2015).
Rapid mass spectrometric conversion of tissue biopsy samples into per-
manent quantitative digital proteome maps. Nat. Med. 21, 407–413.
https://doi.org/10.1038/nm.3807.
31. Coscia, F., Doll, S., Bech, J.M., Schweizer, L., Mund, A., Lengyel, E.,
Lindebjerg, J., Madsen, G.I., Moreira, J.M., and Mann, M. (2020). A
streamlined mass spectrometry–based proteomics workflow for large-
scale FFPE tissue analysis. J. Pathol. 251, 100–112. https://doi.org/10.
1002/path.5420.
32. Cunningham, R.P., and Porat-Shliom, N. (2021). Liver zonation – revisiting
old questions with new technologies. Front. Physiol. 12, 1433. https://doi.
org/10.3389/FPHYS.2021.732929/BIBTEX.
ll
33. Milo, R., Jorgensen, P., Moran, U., Weber, G., and Springer, M. (2010).
BioNumbers—the database of key numbers in molecular and cell biology.
Nucleic Acids Res. 38, D750–D753. https://doi.org/10.1093/NAR/GKP889.
34. Nwosu, A.J., Misal, S.A., Truong, T., Carson, R.H., Webber, K.G.I., Axtell,
N.B., Liang, Y., Johnston, S.M., Virgin, K.L., Smith, E.G., et al. (2022). In-
Depth mass spectrometry-Based proteomics of formalin-Fixed, Paraffin-
Embedded Tissues with a Spatial Resolution of 50–200 mm. J. Proteome
Res. 21, 2237–2245. https://doi.org/10.1021/acs.jproteome.2c00409.
35. Balgley, B.M., Guo, T., Zhao, K., Fang, X., Tavassoli, F.A., and Lee, C.S.
(2009). Evaluation of archival time on shotgun proteomics of formalin-fixed
and paraffin-embedded tissues. J. Proteome Res. 8, 917–925. https://doi.
org/10.1021/pr800503u.
36. Wang, D., Eraslan, B., Wieland, T., Hallström, B., Hopf, T., Zolg, D.P.,
Zecha, J., Asplund, A., Li, L.H., Meng, C., et al. (2019). A deep proteome
and transcriptome abundance atlas of 29 healthy human tissues. Mol.
Syst. Biol. 15, e8503. https://doi.org/10.15252/MSB.20188503.
37. Massoni-Badosa, R., Soler-Vila, P., Aguilar-Fernández, S., Nieto, J.C.,
Elosua-Bayes, M., Marchese, D., Kulis, M., Vilas-Zornoza, A., Bu €hler,
M.M., Rashmi, S., et al. (2022). An atlas of cells in the human tonsil.
https://doi.org/10.1101/2022.06.24.497299.
38. De Silva, N.S., and Klein, U. (2015). Dynamics of B cells in germinal cen-
tres. Nat. Rev. Immunol. 15, 137–148. https://doi.org/10.1038/nri3804.
39. Bankhead, P., Loughrey, M.B., Fernández, J.A., Dombrowski, Y., McArt,
D.G., Dunne, P.D., McQuaid, S., Gray, R.T., Murray, L.J., Coleman, H.G.,
et al. (2017). QuPath: open source software for digital pathology image anal-
ysis. Sci. Rep. 7, 16878. https://doi.org/10.1038/s41598-017-17204-5.
40. Victora, G.D., and Nussenzweig, M.C. (2012). Germinal centers. Annu.
Rev. Immunol. 30, 429–457. https://doi.org/10.1146/annurev-immunol-
020711-075032.
41. Lewis, S.M., Asselin-Labat, M.L., Nguyen, Q., Berthelet, J., Tan, X.,
Wimmer, V.C., Merino, D., Rogers, K.L., and Naik, S.H. (2021). Spatial
omics and multiplexed imaging to explore cancer biology. Nat. Methods
18, 997–1012. https://doi.org/10.1038/s41592-021-01203-6.
42. Merritt, C.R., Ong, G.T., Church, S.E., Barker, K., Danaher, P., Geiss, G.,
Hoang, M., Jung, J., Liang, Y., McKay-Fleisch, J., et al. (2020). Multiplex
digital spatial profiling of proteins and RNA in fixed tissue. Nat.
Biotechnol. 38, 586–599. https://doi.org/10.1038/s41587-020-0472-9.
43. Zhang, B., Wang, J., Wang, X., Zhu, J., Liu, Q., Shi, Z., Chambers, M.C.,
Zimmerman, L.J., Shaddox, K.F., Kim, S., et al. (2014). Proteogenomic
characterization of human colon and rectal cancer. Nature 513,
382–387. https://doi.org/10.1038/nature13438.
44. Derks, J., Leduc, A., Huffman, R.G., Specht, H., Ralser, M., Demichev, V.,
and Slavov, N. (2021). Increasing the throughput of sensitive proteomics
by plexDIA. Nat. Biotechnol. 41, 50–59.
45. Friedrich, C., Schallenberg, S., Kirchner, M., Ziehm, M., Niquet, S., Haji,
M., Beier, C., Neudecker, J., Klauschen, F., and Mertins, P. (2021).
Comprehensive micro-scaled proteome and phosphoproteome charac-
terization of archived retrospective cancer repositories. Nat. Commun.
12, 3576. https://doi.org/10.1038/s41467-021-23855-w.
46. Thielert, M., Itang, C., Ammar, C., Schober, F., Bludau, I., Skowronek, P.,
Wahle, M., Zeng, W.-F., Zhou, X.-X., Brunner, A.-D., et al. (2022). Robust
dimethyl-based multiplex-DIA workflow doubles single-cell proteome
depth via a reference channel Author list. https://doi.org/10.15252/msb.
202211503.
47. Perez-Riverol, Y., Csordas, A., Bai, J., Bernal-Llinares, M., Hewapathirana,
S., Kundu, D.J., Inuganti, A., Griss, J., Mayer, G., Eisenacher, M., et al.
(2019). The PRIDE database and related tools and resources in 2019:
improving support for quantification data. Nucleic Acids Res. 47, D442–
D450. https://doi.org/10.1093/nar/gky1106.
€dler, S.C., Wallmann, G., Metousis, A., Bérouti, M.,
48. Schmacke, N.A., Ma
Harz, H., Leonhardt, H., Mann, M., and Hornung, V. (2023). SPARCS, a
platform for genome-scale CRISPR screening for spatial cellular pheno-
types. https://doi.org/10.1101/2023.06.01.542416.
Cell Systems 14, 1002–1014, November 15, 2023 1013
 ll
49. Frankenfield, A.M., Ni, J., Ahmed, M., and Hao, L. (2022). Protein contam-
inants matter: building universal protein contaminant libraries for DDA and
DIA proteomics. J. Proteome Res. 21, 2104–2113. https://doi.org/10.
1021/acs.jproteome.2c00145.
50. Tyanova, S., Temu, T., Sinitcyn, P., Carlson, A., Hein, M.Y., Geiger, T.,
Mann, M., and Cox, J. (2016). The Perseus computational platform for
1014 Cell Systems 14, 1002–1014, November 15, 2023
Methods
comprehensive analysis of (prote)omics data. Nat. Methods 13,
731–740. https://doi.org/10.1038/nmeth.3901.
51. Cox, J., and Mann, M. (2012). 1D and 2D annotation enrichment: a statis-
tical method integrating quantitative proteomics with complementary
high-throughput data. BMC Bioinformatics 13, S12. https://doi.org/10.
1186/1471-2105-13-S16-S12.
 ll
Methods

STAR+METHODS

KEY RESOURCES TABLE

REAGENT or RESOURCE SOURCE IDENTIFIER

Antibodies
Recombinant Anti-Sodium Potassium Abcam Cat#ab76020;
ATPase antibody [EP1845Y] RRID: AB_1310695
Recombinant Alexa Fluor 647 Anti-CD3D Abcam Cat#ab198937; RRID: AB_2889190
antibody [EP4426]
CD20 Monoclonal Antibody (L26), Alexa Thermo Fisher Scientific Cat# 53-0202-80, RRID: AB_10734357
Fluor 488, eBioscience
Pan Cytokeratin Monoclonal Antibody Thermo Fisher Scientific Cat# 41-9003-80, RRID: AB_11217482
(AE1/AE3), eFluor 570, eBioscience
Biological samples
Murine liver FFPE tissue from C57BL/6 mice The Jackson Laboratory https://www.jax.org/strain/000664
Human tonsil FFPE tissue Institute of Pathology at Charite https://pathologie-ccm.charite.de/
University Hospital
Chemicals, peptides, and recombinant proteins
n-Dodecyl-beta-Maltoside (DDM) Sigma-Aldrich Cat# D4641-500MG
Endoproteinase Lys-C Promega Cat# VA1170
Proteomics grade modified trypsin Promega Cat# V5117
Tris(2-carboxyethyl)phosphine Sigma-Aldrich Cat# C4706-2G
hydrochloride
Acetonitrile (ACN) HPLC-grade VWR Cat# 83640.290
Isopropanol (ISO) Sigma-Aldrich Cat# 1070222511
Triethylammonium bicarbonate pH Merck T7408-100ML
8.5 (TEAB)
Formic acid Merck Millipore Cat# 1.00264.1000
Trifluoroacetic acid Sigma-Aldrich Cat# 96924-250ML-F
2-chloroacetamide Sigma-Aldrich Cat# C0267-100G
EnVision FLEX Target Retrieval Solution Agilent Dako Cat# K8004
High pH (50X)
Microscopy Neo-Clear Sigma-Aldrich Cat# 1.09843.5000
Odyssey Blocking Buffer LI-COR Biosciences Cat# 927-70001
Prolong Diamond antifade mounting Invitrogen Cat# P36961
medium
Aqua Poly Mount Polysciences Europe GmbH Cat# 18606-20
Pepsin solution for antigen retrieval Agilent Dako Cat# S3002
Deposited data
Mass spectrometric raw files This paper ProteomeXchange: PXD042367
List of hepatocyte specific markers 10.1016/j.cmet.2014.11.002 1-s2.0-S1550413114004999-mmc5.xlsx
(Azimifar et al.22)
Software and algorithms
DIA-NN version 1.8.1 Demichev et al.28 https://github.com/vdemichev/DiaNN
R version 4.2.2 The R Project for Statistical https://www.r-project.org/
Computing
Perseus version 1.6.15.0 Tyanova et al.50 https://maxquant.net/perseus/
Leica Laser Microdissection software Leica Microsystems https://www.leica-microsystems.com/
version 8.3.0.08259 products/microscope-software/p/leica-
lmd-software/
(Continued on next page)

Cell Systems 14, 1002–1014.e1–e5, November 15, 2023 e1

 ll
Methods

Continued
REAGENT or RESOURCE SOURCE IDENTIFIER
Zeiss ZEN version 3.7 Carl Zeiss AG https://www.zeiss.com/microscopy/de/
produkte/software/zeiss-zen.html
LAS X software version 3.7.524914 Leica Microsystems https://www.leica-microsystems.com/
products/microscope-software/p/leica-
las-x-ls/
Qupath version 0.4.3 Bankhead et al.39 https://qupath.github.io/
Biological Image Analysis Single Cell Technologies https://single-cell-technologies.com/bias-
Software (BIAS) 2/BioStudies Archive accession number S-
BSST820
Bruker Compass Data Analysis Bruker Daltonik GmbH https://www.bruker.com/en/products-and-
Software version 6.0 solutions/mass-spectrometry/ms-
software.html
Qupath_to_LMD function: Contour This paper https://doi.org/10.5281/zenodo.8414787
export from Qupath to the Leica LMD7
Other
C18 Evotips (EV2013, Evotip Evosep Biosystems https://www.evosep.com/evotip/
Pure, Evosep)
96-well plate Thermo Fisher Scientific https://www.thermofisher.com/order/
catalog/product/de/en/AB1300
384-well low-binding plate Eppendorf Cat# 0030129547
Super PAP-pen liquid Science Services Cat# N71312-N
blocker mini
Cover glass Corning Cat# CLS2980223, #1.5
PPS frame slides Leica Cat# 11600294
PEN glass slides Carl Zeiss Cat# 15350731

RESOURCE AVAILABILITY

Lead contact
Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Fabian
Coscia (fabian.coscia@mdc-berlin.de).

Materials availability
This study did not generate new materials.

Data and code availability

d The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium (http://proteomecentral.
proteomexchange.org) via the PRIDE partner47 with the dataset identifier PXD042367. All data reported in this paper will be
shared by the lead contact upon request.
d The source code for processing the shapes used for laser microdissection has been deposited and is freely available at github.
com/CosciaLab/Qupath_to_LMD. The DOI is listed in the key resources table.
d Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Mouse experiments and organ harvesting

For the mouse liver proteome experiments, 6- 8 weeks old female C57BL/6 mice from Jackson Laboratory were used. C57BL/6 mice
were housed in individually ventilated cages in a specific pathogen-free mouse facility at the Max-Delbru €ck Center for Molecular
Medicine (Berlin, Germany).
For liver excision, anesthetized mice were sacrificed by cervical dislocation, and the livers were removed, rinsed twice in ice-cold
PBS, and transferred to 4% formaldehyde solution for fixation (fixation for at least 24h to 48h). Thereafter, livers were paraffin-
embedded for further histological analyses. The animal experiments were performed in accordance with the United Kingdom Coor-
dinated Committee on Cancer Research (UKCCR) guidelines and were approved by local governmental authorities (Landesamt fu €r
Gesundheit und Soziales Berlin, Germany).

e2 Cell Systems 14, 1002–1014.e1–e5, November 15, 2023

 ll
Methods

Human tissue samples

Tonsil tissues were obtained from two female patients aged 34 and 36. Both presented with recurrent chronic tonsillitis and under-
went bilateral tonsillectomy. The pathological examination revealed hypertrophy and hyperplasia of the lymphoid follicles as well as
enlarged germinal centers. Furthermore, an abundance of collagen fibers within the stroma could be observed. There were no signs
of active inflammation or malignancy.
Resected tonsils were fixed in 10% buffered formalin before gross processing. After overnight fixation, the specimens were weighed,
measured, and macroscopically evaluated. Afterward, the specimens were cut into 5-mm-thick slices. Two representative slices were
embedded in paraffin. The embedded tissue blocks were cut into 2-mm-thick sections, and stained with hematoxylin and eosin for his-
tological examination. The tissue blocks were then stored at room temperature at the archive of the Institute of Pathology at Charité Uni-
versity Hospital, Campus Mitte. The study was performed according to the ethical principles for medical research of the Declaration of
Helsinki and approval was obtained from the Ethics Committee of the Charité University Medical Department in Berlin (EA1/222/21).

METHOD DETAILS

Hematoxylin-eosin staining
Briefly, PEN glass slides (Carl Zeiss, 15350731) were treated with UV light for 1 hour. PPS frame slides (Leica, 11600294) were used
directly for the next steps. FFPE tissue sections were cut with a microtome (5 mm or 10 mm-thick), air dried at 37  C overnight and
heated at 60 C for 10 minutes to facilitate better tissue adhesion. Next, tissue sections were deparaffinized by washing
2x5 minutes in Neo Clear (Sigma Aldrich, 1.09483.5000), followed by a series of 99%, 80% and 70% ethanol for 2 minutes, respec-
tively, and rehydrated by immersing in milliQ water three times. Then, slides were stained in Mayer’s hematoxylin for three minutes
and immersed in tap water for another ten minutes, rinsed in milliQ water and stained with eosin for 30 seconds. Subsequently, the
slides were dehydrated by submerging in 70%, 80%, and 99% ethanol serially. Samples were finally air-dried and stored at RT until
imaging. Before imaging, a cover glass (Corning, CLS2980223, #1.5) was mounted with Aqua Poly Mount medium (Polysciences Eu-
rope GmbH, 18606-20).

Immunofluorescence staining
Following tissue sectioning, mounting on LMD slides and de-paraffinization (described above), two epitope retrieval methods were
compared: heat-induced (HIER) and protease (pepsin)-induced (PIER, 10 min) epitope retrieval. HIER was done by submerging in
EnVision FLEX Target Retrieval Solution High pH solution (diluted to 1X) (Agilent Dako, cat.no. K8004) and heating in a steamer at
95 C for 20 min, and subsequently cooled down in a pre-heated PBS buffer at room temperature for 30 min. Odyssey Blocking Buffer
(LI-COR BioScience, 927-70001) was used for blocking in a humidified chamber for 30 min at room temperature. PIER was done by
applying pepsin (Agilent Dako, cat.no. S3002) on the tissue slide for 5 minutes at 37 C, and immersed in PBS solution to stop the
reaction. The slides were washed two times with PBS and air dried before antibody staining.
For murine liver tissue stains, the primary antibody targeting Na/K-ATPase (stock concentration 0.563 mg/ml, dilution 1:100, Ab-
cam, ab76020) was diluted in Odyssey Blocking Buffer and incubated overnight at 4  C in a humidified chamber. Next, tissue spec-
imens were washed 3x in PBS and secondary antibodies for the visualization of Na/K-ATPase (Alexa Fluor 488 donkey anti-rabbit,
stock concentration 2 mg/ml, dilution 1:250, A32790, Invitrogen) were diluted in Odyssey Blocking Buffer and applied for 1 hour at
room temperature in the dark. After staining, slides were washed 3x in PBS, counterstained by Hoechst (dilution 1:1000 in PBS,
Thermo Fisher Scientific, #62249) for 10 minutes and followed by three washes in PBS and two washes in milliQ. Before imaging,
a cover glass (#1.5) was mounted using ProLong Diamond anti-fade mounting medium (Thermo Fisher Scientific, P36961).
For tonsil tissue stains following tissue sectioning, mounting on LMD slides and de-paraffinization, the tissues were subjected to
heat-induced epitope retrieval as described above. Odyssey Blocking Buffer was used for blocking in a humidified chamber for
30 min at room temperature. Next, conjugated primary antibodies targeting CD20 (stock concentration 0.5mg/ml, dilution 1:50, Ther-
mofisher, 53-0202-80, Alexa Fluor 488), CD3 (stock concentration 0.5 mg/ml, dilution 1:100, Abcam, ab198937, Alexa Fluor 647), and
pan-cytokeratin (stock concentration 0.2 mg/ml, dilution 1:100, Thermofisher, 41-9003-80, eFluor 570) were diluted in Odyssey
Blocking Buffer (and incubated overnight at 4  C in a humidified chamber. Tissue specimens were washed 4x in PBS, counterstained
by Hoechst (dilution 1:1000 in PBS, Thermo Fisher Scientific, 62249) for 10 minutes, washed 4x in PBS and 2x in milliQ water. Sub-
sequently, the slides were dehydrated by submerging them in 70%, 80%, and 99% ethanol serially. Before imaging, a cover glass
was mounted with ProLong Diamond anti-fade mounting medium.

High-resolution microscopy
Images of immunofluorescence-labeled tonsil tissue sections were acquired using an Axioscan 7 system (Zeiss), equipped with
wide-field optics, a Plan-A photochromat 10x/0.45 M27 objective and a quadruple-band filter set for Alexa fluorescent dyes. The
wide-field acquisition was performed using the Colibri 7 LED light source and an AxioCam 712m camera. Images were obtained auto-
matically with Zeiss ZEN 3.7 (blue edition) at non-saturating conditions (16-bit dynamic range).
Images of immunofluorescently-labeled and hematoxylin-eosin stained murine liver tissue sections were acquired on the Leica
LMD7 system using the LAS X software (version 3.7.524914, Leica Microsystems), an HC PL FLUOTAR 10x/0.32 DRY objective
and the DF7000T camera. The microscope was equipped with the following filter sets: LMD-Dapi, LMD-Cy3, LMD - Alexa594,

Cell Systems 14, 1002–1014.e1–e5, November 15, 2023 e3

 ll
Methods

LMD-YFP, LMD-CY5, LMD-BGR, GFP. After imaging, the cover glass was finally removed through gentle agitation of the slide in PBS.
After one dip in milliQ water for salt removal, the slide was air-dried and stored at 4 C until laser microdissection.

Image analysis and contour export for laser microdissection

Image analysis was performed in QuPath (version 0.4.3) and BIAS (BioStudies Archive accession number S-BSST820). Annotations
of different regions of interest were manually created in QuPath after image analysis. The assignment of three reference points (x-y
coordinates) is required for precise contour transfer between the screening and laser microdissection microscopes. Contours and
reference points were exported in a geojson format and translated into the .XML format compatible with the Leica LMD7 software.
The code for processing the shapes is available at github.com/CosciaLab/Qupath_to_LMD, it uses geopandas (Version 0.12.2) and
py-lmd48 (Version 1.0.0).

Laser microdissection
We used the Leica LMD 7 system and Leica Laser Microdissection V 8.3.0.08259 software for the collection of tissue contours. De-
pending on the contour size, tissue was cut with a 20x or 63x objective in fluorescence or brightfield mode. The following laser set-
tings were used for the 20x objective (HC PL FL L 20x/0.40 CORR): power 56, aperture 1, speed 15, middle pulse count 1, final pulse
-1, head current 37 – 45% (depending on tissue type and section thickness), pulse frequency 801 and offset 101. For the 63x objec-
tive (HC PL FLUOTAR L 63x/0.70 CORR XT), the settings were: power 59, aperture 1, speed 50, middle pulse count 1, final pulse -1,
head current 45-50%, pulse frequency 3000, offset 101.
Contours were cut and sorted into a low-binding 384-well plate (Eppendorf 0030129547) configured over the ‘universal holder’
function with one empty well between samples.

Sample preparation for LC-MS analysis

A detailed sample preparation protocol is provided in the supplemental information (Methods S1). Tissue samples were collected by
manual cutting or by automated cutting after contour import into low-binding 384-well plates. For mouse liver tissue samples (5- or
10-mm-thick sections cut with a microtome), regions of 600 mm2 - 100,000 mm2 were collected. To concentrate tissue pieces at the
bottom of each well after LMD collection, 15 ml of acetonitrile was added to each well, briefly vortexed and vacuum dried (15min at
60 C). Another well inspection is recommended before proteomics sample preparation to ensure high collection efficiency.
We tested three different protocols, DDM-based, ACN-based and a combination of DDM and ACN. The lysis buffer for the DDM-
based protocol consisted of 0.1% DDM, 5mM TCEP, 20mM CAA and 0.1M TEAB in water. 2ml of lysis buffer was added to each sam-
ple well using the MANTIS Liquid Dispenser (Formulatrix, V3.3 ACC RFID, software version 4.7.5) and the high-volume diaphragm
chips (Formulatrix, cat.no. 233128). The plate was closed with a PCR ComfortLid (Hamilton), and heated at 95 C for 60 minutes.
Then, samples were shortly cooled down, and 1ml of LysC was added (prediluted in water to 2 ng/ml) and digested for minimum 2
hours at 37 C in the thermal cycler (50 C lid temperature). Subsequently, 1ml of trypsin was added (prediluted in water to 2 ng/ml)
and incubated overnight at 37  C in the thermal cycler. The next day, digestion was stopped by adding trifluoroacetic acid (TFA, final
concentration 1% v/v), and samples were vacuum dried before peptide clean-up.
For the ACN-based protocol, the lysis buffer consisted of 5mM TCEP, 20mM CAA, 0.1M TEAB diluted in water. 2 ml of lysis buffer
was added to each sample well, the plate was closed with PCR ComfortLid, and heated at 95 C for 60 minutes in a thermal cycler
(Bio-Rad S1000 with 384-well reaction module) at a constant lid temperature of 110 C. Next, 1 ul of 100% ACN was added to each
well, and the plate was incubated for an additional 60 min in a thermal cycler at 75 C at a constant lid temperature of 110 C. Then,
samples were shortly cooled down, and 1 ml of LysC was added (prediluted with 0.1M TEAB, 30% ACN water to 2 ng/ml) and digested
for 2 hours at 37 C in the thermal cycler (50 C lid temperature). Subsequently, 1 ml of trypsin was added (prediluted with 0.1M TEAB,
10% ACN water to 2 ng/ml) and incubated overnight at 37  C in the thermal cycler. The next day, digestion was stopped by adding
trifluoroacetic acid (TFA, final concentration 1% v/v), and samples were vacuum dried before peptide clean-up.
For the combined DDM/ACN-based protocol, the lysis buffer for the DDM-based protocol consisted of 0.1% DDM, 5mM TCEP,
20mM CAA and 0.1M TEAB in water. 2ml of lysis buffer was added to each sample well using the MANTIS Liquid Dispenser and the
high-volume diaphragm chips. The plate was closed with a PCR ComfortLid, and heated at 95 C for 60 minutes. Then, samples were
shortly cooled down, and 1 ml of LysC was added (2 ng/ml in 0.1M TEAB [pH 8.5] and 30% ACN in milliQ water) and digested for min-
imum 2 hours at 37 C in the thermal cycler (50 C lid temperature). Subsequently, 1 ml of trypsin was added (2 ng/ml containing 10%
ACN and 0.1M TEAB [pH 8.5] in milliQ water.) and incubated overnight at 37  C in the thermal cycler. The next day, digestion was
stopped by adding trifluoroacetic acid (TFA, final concentration 1% v/v), and samples were vacuum dried before peptide clean-up.

Peptide clean-up with C-18 tips

Evotip (Evosep, Odense, Denmark) based peptide clean-up was performed as recommended by the manufacturer. Briefly, 20 ul of
buffer B (99.9% ACN, 0.1% FA) was added to each C-18 tip (EV2013, Evotip Pure, Evosep) and centrifuged at 700 rpm for 1 minute.
Then, 20 ul of buffer A (99.9% water, 0.1% FA) was added from the top of each C-18 tip, activated in isopropanol for 20 seconds and
centrifuged again at 700 rpm for 1 minute. Digested tissue samples were then loaded onto Evotips, washed once with 20 ul buffer A
and finally eluted with 20 ul buffer B to a 96-well plate (Thermo Fisher Scientific, AB1300), and vacuum dried (15 min at 60 C).

e4 Cell Systems 14, 1002–1014.e1–e5, November 15, 2023

 ll
Methods

Samples were stored at 20  C until liquid chromatography–mass spectrometry (LC–MS) analysis. For LC-MS analysis, 4.2 ml of MS
loading buffer (3% acetonitrile, 0.1% TFA in water) was added, the plate was vortexed for 10 seconds and centrifuged for 1 minute at
700g. 4 ml were finally injected into the mass spectrometer.

Liquid chromatography–mass spectrometry (LC – MS) analysis

LC-MS analysis was performed with an EASYnLC-1200 system (Thermo Fisher Scientific) connected to a trapped ion mobility spec-
trometry quadruple time-of-flight mass spectrometer (timsTOF SCP and timsTOF Pro2, Bruker Daltonik) with a nano-electrospray ion
source (CaptiveSpray, Bruker Daltonik). The autosampler was configured to pick samples from 384- and 96-well plates.
Peptides were loaded on a 20-cm home-packed HPLC column (75-mm inner diameter packed with 1.9-mm ReproSil-Pur C18-AQ
silica beads, Dr. Maisch).
Peptides were separated using a linear gradient from 7-30% buffer B (0.1% formic acid and 90% ACN in LC-MS grade water) in
14 minutes, followed by an increase to 60% for 1 minute and a 1.5-minute wash in 90% buffer B at 250 nl min-1. Buffer A consisted of
0.1% formic acid in LC-MS grade water. The total gradient length was 21 minutes. A column oven was used to keep the column tem-
perature constant at 40 C.
For dia-PASEF analysis, we used a dia-PASEF method with 8 dia-PASEF scans separated into 3 ion mobility windows per scan
covering a 400-1000 m/z range by 25 Th windows and an ion mobility range from 0.64 to 1.37 Vs cm-2. The mass spectrometer
was operated in high sensitivity mode, with an accumulation and ramp time at 100 ms, capillary voltage set to 1750V and the collision
energy as a linear ramp from 20 eV at 1/K0 = 0.6 Vs cm-2 to 59 eV at 1/K0 = 1.6 Vs cm-2. The collision energy was ramped linearly as a
function of ion mobility from 59 eV at 1/K0 = 1.6 V s cm-2 to 20 eV at 1/K0 = 0.6 V s cm-2.

Proteomics raw data analysis

We used DIA-NN28 (version 1.8.1) for dia-PASEF raw file analysis and spectral library generation.
Spectral library generation
For spectral library generation to analyze dia-PASEF data, human and mouse FASTA files were downloaded from Uniprot (2022
release, UP000000589_10090 Mus Musculus, UP000005640_9606, downloaded on April 10th and April 8th 2022 respectively).
DIA-NN in silico predicted libraries were generated by providing the human or mouse FASTA file and frequently found contaminants49
(mouse + mouse tissue contaminants, or human + universal contaminants). Deep learning-based spectra, RTs and IMs prediction
were enabled for the appropriate mass range of 300-1200 m/z. N-terminal M excision was enabled and cysteine carbamidomethy-
lation was enabled as a fixed modification. A maximum of 2 missed cleavages was allowed, and the precursor charge set to 2 - 4. For
the generation of project-specific refined libraries, in-silico-generated mouse and human libraries were used to search 20-50 raw files
of high-sample amounts, such that the optimal total ion current was reached (see Figures 2 and 3). The refined murine liver library
consisted of 68,006 precursors, 61,554 elution groups and 8,225 protein groups. The refined human tonsil library consisted of
47,999 precursors, 44,675 elution groups and 8,137 protein groups.
Search of dia-PASEF raw files with refined libraries
DIA-NN was operated in the default mode with minor adjustments. Briefly, MS1 and MS2 accuracies were set to 15.0, scan windows
to 0 (assignment by DIA-NN), isotopologues were enabled, no MBR for project-specific DIA-NN-refined libraries, heuristic protein
inference and no shared spectra. Enabling MBR for refined libraries improved data completeness of single-cell samples (Figure
S2G). No difference was observed for higher input samples. Proteins were inferred from genes, neural network classifier was set
to single-pass mode, quantification strategy as ‘Robust LC (high precision)’. Cross-run normalization was set to ‘RT-dependent’, li-
brary generation as ‘smart profiling’, speed and Ram usage as ‘optimal results’.

QUANTIFICATION AND STATISTICAL ANALYSIS

Proteomics data analysis was performed with Perseus50 (version 1.6.15.0) and within the R environment (https://www.r-project.org/,
version 4.2.2) with the following packages: ggplot2 (v3.4.2), FactoMineR (v2.8), factoextra (v 1.0.7.999), plotly (v4.10.1), reshape2
(v1.4.4), viridis (v0.6.3), UpSetR (v1.4.0). For differential expression analysis (t-test or ANOVA, Figures 3 and 4), data were filtered
to keep only proteins with 70% non-missing data in at least one group. Missing values were imputed based on a normal distribution
(width = 0.3, downshift = 1.8) before statistical testing. For multi-sample (ANOVA) or pairwise proteomic comparisons (two-sided un-
paired t-test), a permutation-based FDR of 5% was applied to correct for multiple hypothesis testing. Principal component analysis
was performed in R (see packages above). 1D pathway enrichment analysis51 (Figure 4I) was done in Perseus based on the KEGG
(https://www.genome.jp/kegg/) and Reactome Pathway Database (reactome.org), using a Benjamini-Hochberg FDR cut-off of 0.05.
The minimum category size was set to 5.

Cell Systems 14, 1002–1014.e1–e5, November 15, 2023 e5